Tilburg University
Audiovisual prosody and feeling of knowing
Swerts, Marc; Krahmer, Emiel
Published in:
Journal of Memory and Language
Publication date:
2005
Link to publication
Citation for published version (APA):
Swerts, M. G. J., & Krahmer, E. J. (2005). Audiovisual prosody and feeling of knowing. Journal of Memory and
Language, 53(1), 81-94.
General rights
Copyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright owners
and it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.
- Users may download and print one copy of any publication from the public portal for the purpose of private study or research
- You may not further distribute the material or use it for any profit-making activity or commercial gain
- You may freely distribute the URL identifying the publication in the public portal
Take down policy
If you believe that this document breaches copyright, please contact us providing details, and we will remove access to the work immediately
and investigate your claim.
Download date: 18. jun. 2017
Journal of Memory and Language 53 (2005) 81–94
www.elsevier.com/locate/jml
Audiovisual prosody and feeling of knowing
Marc Swerts, Emiel Krahmer ¤
Communication and Cognition, Faculty of Arts, Tilburg University, P.O. Box 90153, NL-5000 LE Tilburg, The Netherlands
Received 9 July 2004; revision received 2 February 2005
Available online 16 March 2005
Abstract
This paper describes two experiments on the role of audiovisual prosody for signalling and detecting meta-cognitive
information in question answering. The Wrst study consists of an experiment, in which participants are asked factual
questions in a conversational setting, while they are being Wlmed. Statistical analyses bring to light that the speakers’
Feeling of Knowing (FOK) is cued by a number of visual and verbal properties. It appears that answers tend to have a
higher number of marked auditory and visual cues, including divergences from the neutral facial expression, when the
FOK score is low, while the reverse is true for non-answers. The second study is a perception experiment, in which a
selection of the utterances from the Wrst study is presented to participants in one of three conditions: vision only, sound
only, or vision + sound. Results reveal that human observers can reliably distinguish high FOK responses from low
FOK responses in all three conditions, but that answers are easier than non-answers, and that a bimodal presentation of
the stimuli is easier than the unimodal counterparts.
 2005 Elsevier Inc. All rights reserved.
Keywords: Question answering; Feeling of knowing; Feeling of another’s knowing; Tip of the tongue; Audiovisual prosody; Facial
expressions; Speech production; Speech perception
Speakers are not always equally conWdent about or
committed to what they are saying. When asked a question, for instance, they can be certain or rather doubtful
about the correctness of their answer, and they may be
unable to respond at all, even though in some cases it
might feel as if the answer lies on the tip of the tongue. It
has been shown that such meta-cognitive aspects of
question answering are reXected in the way speakers
‘package’ their utterances prosodically (Brennan & Williams, 1995; Smith & Clark, 1993), where prosody can
broadly be deWned as the whole gamut of features that
*
Corresponding author. Fax: +31 13 4663110.
E-mail addresses: [email protected] (M.
[email protected] (E. Krahmer).
Swerts),
do not determine what speakers say, but rather how they
say it (Bolinger, 1985; Cruttenden, 1986; Ladd, 1996).
The prosodic cues may be helpful to addressees, because
they can use them to determine whether or not they
should interpret incoming sentences with a grain of salt
(cf. Sabbagh & Baldwin, 2001). Sofar, research has
focussed primarily on analyses of verbal features, such as
particular intonation patterns, that are encoded in the
speech signal itself. Visual cues, such as a wide range of
gaze patterns, gestures, and facial expressions, are a natural and important ingredient of communication as well.
It seems a reasonable hypothesis that these might be
informative just as verbal features are, though this issue
is largely unexplored. The current paper tries to Wll this
gap by looking at possible signals of meta-cognitive
speaker states on the basis of combinations of verbal
0749-596X/$ - see front matter  2005 Elsevier Inc. All rights reserved.
doi:10.1016/j.jml.2005.02.003
82
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
and visual features, for which we will use the term audiovisual prosody (following, e.g., Barkhuysen, Krahmer, &
Swerts, 2005; Krahmer & Swerts, 2004; Munhall, Jones,
Callan, Kuratate, & Vatikiotis-Bateson, 2004; Srinivasan
& Massaro, 2003). Since our work builds on insights of
past studies of memory processes in question answering
situations (e.g., Brennan & Williams, 1995; Smith &
Clark, 1993), we Wrst discuss such prior work in more
detail.
In traditional models of memory, answering factual
questions (e.g., “Who wrote Hamlet?”) proceeds as follows: when asked a question, a person searches for the
answer in memory. If the answer is found, it will be produced, else the person signals search failure by providing
a non-answer such as “I don’t know.” A major deWcit of
such simple models is that they leave unspeciWed for how
long a person should search for an answer and what triggers the decision to produce a non-answer. For these reasons current models of memory often supplement the
searching component with a meta-cognitive monitoring
component, which continuously assesses a person’s
knowledge. For each potential answer that the searching
component retrieves, the monitoring component estimates the conWdence in the answer’s correctness. If the
conWdence is high enough, the answer can be produced.
If not, the search component can look for more conWdent answers, provided the meta-memory assessment
indicates that a proper answer may still be found (see,
e.g., Andersen, 1999; Koriat, 1993; Nelson, 1996).
Smith and Clark (1993) emphasize that in a conversational, every day setting, question answering is a social
process, which involves information exchange as well as
self-presentation. To account for this, Smith and Clark
point out that a further component should be added to a
model for question answering, namely a responding component, which uses the output of both search and monitoring to provide feedback to the questioner. For
instance, when a person responds to a question with an
answer that has a relatively low conWdence score, the
person can signal this uncertainty in his or her response;
if the answer would turn out to be incorrect later on, the
speaker might save face and look less foolish since it was
suggested that there was little conWdence in the response
(GoVman, 1967). Smith and Clark (1993) note that signalling uncertainty of an answer can be done, for
instance, via explicit commentary, using linguistic hedges
such as “I guess” or “perhaps,” or, more implicitly, by
using a rising, question-like intonation (e.g., Bolinger,
1985; Morgan, 1953). Similarly, speakers may want to
account for delays using Wllers such as “uh” or “uhm”
(e.g., Clark & Fox Tree, 2002), when the search has not
resulted in an answer yet, even though the meta-memory
assessment indicates that an acceptable answer may still
be found after a prolonged search.
The resulting 3-fold model of question answering
(searching, monitoring, and responding) yields a number
of testable predictions: if a speaker Wnds a question diYcult, this will result in a longer delay and correspondingly more Wllers; and this will also result in more
‘marked’ answer behavior (hedges, and paralinguistic
cues such as rising intonation and Wllers). To test these
predictions, Smith and Clark (1993) use Hart’s (1965)
experimental method, a common technique to elicit
meta-memory judgements. In this method participants
are Wrst asked a series of factual questions in a conversational setting. Then they are asked to rate for each question their feeling that they would recognize the correct
answer [i.e., their Feeling of Knowing (FOK)]. Finally
participants take the actual recognition test, i.e., a multiple-choice test in which all the original questions are presented once more. The interesting thing is that this
method provides a handle on the Tip of the Tongue
(TOT) phenomenon where a speaker is unable to produce an answer, but has the “feeling of knowing” it, perhaps because there was partial access to related
information or because the question (cue) seemed familiar (e.g., Koriat, 1993; Reder & Ritter, 1992). Using this
paradigm, Smith and Clark (1993) found that there was
a signiWcant diVerence between two types of responses,
answers and non-answers: the lower a speaker’s FOK,
the slower the answers, but the faster the non-answers. In
addition, the lower the FOK, the more often people
answered with rising intonation, and added Wllers,
explicit linguistic hedges and other face-saving comments. It is worth noting that high FOK answers and
low FOK non-answers are similar in that both express
some form of speaker certainty: in the case of a high
FOK answer the speaker is relatively certain of the
answer, while a low FOK non-answer indicates that the
speaker is relatively certain that he or she does not know
the answer. In both cases, the speaker will not embark on
a longer memory search, and the reply will tend to contain few paralinguistic cues.
The question naturally arises whether other people
can reliably determine the FOK of a speaker. To test
this, Brennan and Williams (1995) Wrst performed an
experiment which replicated Smith and Clark’s earlier
Wndings, and subsequently made a selection of the speakers’ responses which were presented to listeners. These
were tested on their Feeling of Another’s Knowing
(FOAK) to see if meta-cognitive information was reliably conveyed by the surface form of responses. The
results for answers showed that rising intonation and
longer latencies led to lower FOAK scores, whereas for
non-answers longer latencies led to higher FOAK scores.
Brennan and Williams (1995) conclude that listeners can
indeed interpret the meta-cognitive information that
speakers display about their states of knowledge in question answering.
Even though they stress the interactive and social
nature of question answering, both Smith and Clark
(1993) and Brennan and Williams (1995) are only
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
concerned with linguistic and prosodic cues to meta-cognition. Arguably, the speaker cues for meta-cognition
that they consider in their respective studies are obvious
ones, but they are not fully representative of the richness
of face-to-face communication, in which gestures and
facial expressions play important roles as well. The only
study we are aware of in which FOAK judgements were
made by people that could actually see respondents
answering questions is Jameson, Nelson, Leonesio, and
Narens (1993), but in that study the respondents had to
type their answers and were not allowed to speak.
There is ample evidence that visual information is a
crucial element at various levels of face-to-face conversation. A number of studies have shown that visual cues
are beneWcial for the understandability of speech (e.g.,
Jordan & Sergeant, 2000; Sumby & Pollack, 1954). During face-to-face communication, people pay a lot of
attention to gestures (e.g., Kendon, 1994; Krauss, Chen,
& Chawla, 1996; McNeill, 1992) and facial expressions
such as nodding, smiling, and gazing (e.g., Clark &
Krych, 2004): such visual cues help to structure the interaction, for instance, by facilitating turn taking, or they
support/supplement the information a speaker wants to
convey. A number of recent studies also link gestures
and other non-verbal behaviors to meta-cognitive processes (e.g., Frick-Horbury, 2002; Lawrence, Myerson,
Ooknk, & Abrams, 2001; Morsella & Krauss, 2004).
Morsella and Krauss (2004), for instance, explicitly
relate the use of gestures to the TOT phenomenon. They
point out that mental representations are transient and
that retrieving them is eVortful (Farah, 2000; cf. also
Kikyo, Ohki, & Miyashita, 2002). As a result, holding
them in mind for lengthy intervals (as can occur during
question answering) may be diYcult; gestures, they
argue, facilitate speech production more directly, since
they continuously activate the prelinguistic sensorimotor
features that are part of semantic representations of target words (Morsella & Krauss, 2004 refer to this as the
Gestural Feedback Model). Besides gestures, facial cues
can be indicative of meta-cognitive processes as well.
Goodwin and Goodwin (1986) (see also Clark, 1996)
already discussed the communicative relevance of the socalled “thinking face”: it often happens that, during
search, a respondent turns away from the questioner
with a distant look in the eyes in a stereotypical facial
gesture of someone thinking hard. In a similar vein, it
has been argued that gaze may be indicative of memory
processes (e.g., Glenberg, Schroeder, & Robertson, 1998).
Given such earlier work which suggests that visual
information may be a cue for meta-cognition, it is interesting to see whether speakers produce visual cues correlating with their feeling of knowing and whether such
cues can be perceived by others and used to estimate the
feeling of another’s knowing. Additionally, an audiovisual study of meta-cognitive information is relevant from
the view point of multisensory perception. In this Weld of
83
research, one open question is what the relative contributions are of visual and auditory cues in audiovisual perception and how the two groups of cues interact. On the
basis of earlier studies, it can be conjectured that the relative importance of auditory and visual cues diVers for
diVerent aspects of communication. Information presentation, for instance, is traditionally studied from an auditory perspective. At the basic level of understandability
the acoustic signal usually provides suYcient information
(see, e.g., Kohlrausch & van de Par, 2005). Visual cues
tend to inXuence understandability in, for instance,
adverse acoustical conditions (e.g., Benoît, Guiard-Marigny, Le GoV, & Adjoudani, 1996; Jordan & Sergeant,
2000; Sumby & Pollack, 1954) or in the case of incongruencies (as in the well-known McGurk eVect, where
an auditory /ba/ combined with a visual /ga/ is perceived
as /da/ by most subjects, McGurk & MacDonald, 1976).
At the level of prosody a similar dominance of auditory
cues can be observed. For instance, a number of prosodic
studies that focussed on audiovisual signalling of prominence (indicating which words in an utterance are important) found that, even though facial cues such as head
nods and eyebrow movements contribute to perceived
prominence, the auditory cues (i.e., pitch accents) dominate (e.g, Keating et al., 2003; Krahmer & Swerts, 2004;
Swerts & Krahmer, 2004). In general, the vast majority of
studies have concentrated on verbal prosody, while only
recently an interest in audiovisual prosody seems to
emerge. When looking at emotion, on the other hand, it
appears that visual cues (in particular facial expressions)
are judged to be more important than auditory cues such
as voice information (e.g., Bugenthal, Kaswan, Love, &
Fox, 1970; Hess, Kappas, & Scherer, 1988; Mehrabian &
Ferris, 1967; Walker & Grolnick, 1983), and it may be
noted that emotion research initially focussed on facial
perception (see, for instance, the work of Paul Ekman
and colleagues, e.g., Ekman, 1999). Interestingly, various
more recent studies have shown that both modalities
inXuence emotion perception of incongruent stimuli,
where the visual channel may oVer one emotion (‘happy’)
and the auditory channel another (‘sad’) (e.g., de Gelder
& Vroomen, 2000; Massaro & Egan, 1996). Meta-cognitive information during question answering is an interesting candidate to investigate from this audiovisual
perspective, since it shares some characteristics with both
emotion (in that it reXects a speakers’ internal state) and
with information presentation (in the meta-cognitive signalling in the actual response).
In this paper, we describe two experiments which
study the role of audiovisual prosody for the production
and perception of meta-cognitive information in question answering. We will be particularly interested in combinations of auditory cues (rising intonation, Wllers, and
unWlled pauses) with visual cues (such as gaze, eyebrow
movements, and smiles). For Experiment 1, we use
Hart’s (1965) paradigm to elicit answers and non-
84
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
answers with diVerent FOK scores, which we then analyze in terms of a number of audiovisual prosodic features. The study by Brennan and Williams (1995) focused
on auditory cues alone, while the goal of Experiment 2 is
to explore whether observers of speakers’ responses (collected in Experiment 1) are able to guess these speakers’
FOK scores on the basis of visual cues as well. In particular, we are interested in whether a bimodal presentation
of stimuli leads to better FOK predictions than the unimodal components in isolation. While we expect that we
get the best performance for bimodal stimuli, it is an
interesting empirical question whether the auditory or
the visual features from the unimodal stimuli are more
informative for FOK predictions.
Experiment 1
Method
Participants
Twenty participants (11 male, 9 female), colleagues
and students from Tilburg University, between 20 and
50 years old, participated as speakers in the experiment
on a voluntary basis. None of the participants objected
to being recorded, and all granted usage of their data for
research and educational purposes.
Materials
The materials consisted of a series of factual questions, taken in part from a Dutch version of the “Wechsler Adult Intelligence Scale” (WAIS), an intelligence test
for adults. We selected only those questions which would
trigger a one-word response (e.g., Who wrote Hamlet?
What is the capital of Switzerland?), and added a supplementary list of questions from the game Trivial Pursuit.
The 40 questions in total covered a wide range of topics,
like literature, sports, history, etc. The list of questions
was presented to participants in one of two random
orders (see Appendix A).
Procedure
Following Hart (1965) and Smith and Clark (1993),
the experimental procedure consisted of three stages. In
the Wrst stage, the 40 questions were posed by the experimenter, and the responses by the participant were Wlmed
(front view of head). As in Smith and Clark (1993), participants could not see the experimenter and they were told
that this was to avoid the participants from picking up
any potential cues the experimenter might give about the
correctness of their answers. Participants were told that
the experimenter could see them via the digital camera,
which was connected to a monitor behind a screen. The
experimenter asked the series of 40 questions one by one,
and the pace of the experiment was determined by the participant. As an example, here are Wve responses (translated
from Dutch) to the question about the name of the person
who drew the pictures in “Jip en Janneke,” a famous
Dutch book for children written by Annie M.G. Schmidt:
(a)
(b)
(c)
(d)
(e)
Fiep Westendorp
uh Fiep Westen-(short pause)-dorp
(short pause) Isn’t that Annie M.G. Schmidt?
no idea
uh the writer is Fiep Westendorp, but the drawings? No, I don’t know
The example shows cases of correct answers, which
could be Xuent (a) or hesitant (b), an incorrect answer
(c), and a simple (d) and complex (e) case of a nonanswer.
In the second stage, the same sequence of questions
was again presented to the same participants, but now
they had to express on a 7-point scale how sure they
were that they would recognize the correct answer if they
would have to Wnd it in a multiple-choice test, with 7
meaning “deWnitely yes” and 1 “absolutely not.” The
third stage was a paper-and-pencil test in which the same
sequence of questions was now presented in a multiplechoice in which the correct answer was mixed with three
plausible alternatives. For instance, the question “What
is the capital of Switzerland?” listed Bern (correct) with
three other large Swiss cities: Zürich, Genève, and Basel.
Participants were unaware of the real purpose of the
study, but were told that its goal was to learn more
about the average complexity of a series of questions
which could be used in future psycholinguistic research.
They were warned beforehand that questions were
expected to vary in degree of diYculty. To encourage
them to do their best and guess the correct answer in
case of uncertainty, the experiment incorporated an element of competition, where the ‘winner’ (i.e., the person
with the highest number of correct answers in the Wrst
test) got a small reward (a book token).
Labelling, annotation
All utterances from the Wrst test (800 in total) were
transcribed orthographically and manually labelled
regarding a number of auditory and visual features by
four independent transcribers. The labelling was based
on perceptual judgements and features were only
marked when clearly present.
Regarding verbal cues, we labelled the presence or
absence of the following features:
Delay:Whether a speaker responded immediately, or
took some time to respond.
High intonation:Whether a speaker’s utterance ended
in a high or a low boundary tone.
Fillers:Whether the utterance contained one or more
Wllers, or whether these were absent.
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
For simplicity’s sake, all three verbal features were
treated in a binary way. A clearly perceivable pause
before speaking was labelled as a delay while an immediate response was not. We did not measure precise delay
durations as Smith and Clark (1993) and Brennan and
Williams (1995) did. Concerning high intonation, note
that we did not attempt to isolate question intonation, as
it turned out to be diYcult to consistently diVerentiate
‘real’ question intonation from list intonation. Finally,
for the labelling of Wllers we did not diVerentiate between
‘uh,’ ‘uhm’ or ‘mm,’ since the results of the perception
study of Brennan and Williams (1995) only revealed
minor diVerences between them.
In addition to these categorical variables, we counted
the number of words spoken in the utterance, where
family names, like Elvis Presley, were considered to be
one word. This count provides a rough measure for the
amount and length of linguistic hedges (such as “I
guess”) used by speakers, where we did not diVerentiate
between the kind of explicit commentary used.
As to the visual cues, we labelled the presence or
absence of the following features:
Eyebrow movement:If one or more eyebrows departed
from neutral position during the utterance.
Smile:If the speaker smiled (even silently) during the
response.
Low Gaze:Whether a speaker looked downward during the response.
High Gaze:Whether a speaker looked upward during
the response.
Diverted Gaze:Whether a speaker looked away from
the camera (to the left or the right) during the
response.
Funny face:Whether the speaker produced a marked
facial expression, which diverted from a neutral
expression, during the response.
Note that an utterance may contain multiple
marked features. In addition to the isolated features
mentioned above, we also counted the number of
diVerent gaze acts, i.e., combinations of high, low or
diverted gaze. In the literature on facial expressions,
the features gaze (Argyle & Cook, 1976; Gale & Monk,
2000; GriYn, 2001; Glenberg et al., 1998), smiling
(Kraut & Johnson, 1979), and brow movements
(Ekman et al., 1979; Krahmer & Swerts, 2004) are mentioned as informative cues in spoken interaction, and a
Wrst inspection of the recordings revealed that they
occur relatively frequently in the current data set. The
expression we, for want of a better term, called “funny
face” can be seen as speciWc instance of Goodwin and
Goodwin’s (1986) thinking face, explicitly suggesting
uncertainty. Fig. 1 contains some representative stills
for each of the visual features. The visual features are
roughly comparable with Action Units (AUs)
85
described by Ekman and Friesen (1978), though there
is not necessarily a one-to-one mapping to these
Action Units. The action units form the building
blocks of Ekman and Friesen’s Facial Action Coding
System which builds on the assumption that facial
actions can be described in terms of muscular action.
The basic action units correspond with single muscle
activities and more complex facial expressions can be
described using these atomic building blocks. Fig. 1, in
particular, displays examples of marked settings of
smile (AU 12, AU 13), gaze (AU 61–AU 64), and eyebrow raising (AU 1, AU 2). Funny faces typically consist of a combination of AUs such as lip corner
depression (AU 15), lip stretching (AU 20) or lip pressing (AU 24), possibly combined with eye widening
(AU 5), and eyebrow movement as well.
The labelling procedure was as follows. On the basis
of a preliminary labelling protocol, utterances from
two speakers were labelled collectively. For most cues
the labelling was unproblematic, and for the few more
diYcult cases, mostly involving gaze, a consensus labelling was reached after discussion. This Wrst phase
resulted in an explicit labelling protocol, after which
the labellers proceeded individually by labelling a
selection of the cues in the recordings from the remaining 18 speakers. Only the feature gaze was annotated
by two persons who labelled the diVerent gaze acts in
an utterance (low, high, and diverted) in isolation and
resolved mismatches via discussion. All features were
annotated independently from the FOK score to avoid
circularity. Similarly, annotators only looked at
answers without taking the question context into
account, so that the answer’s (in)correctness could not
bias the annotator.
Statistical analyses
Our statistical procedure closely follows the ones
proposed by Smith and Clark (1993) and Brennan and
Williams (1995). We obtained 800 responses, 40 from
each of the 20 participants. These responses are not
independent so that analyses across the entire data set
are inappropriate. Therefore, the following tests are
always based on individual analyses per speaker. We
computed correlation coeYcients for each subject individually, transformed the correlations into Fischer’s zr
scores, and tested the average zr score against zero. The
average zr scores were then transformed back into correlations for reporting in this article. Similarly, when
comparing means, we computed a mean for each
speaker, and used these composite scores in our analyses. In any individual analysis, we did not include any
participant for whom we could not compute an individual correlation or mean, so some of our statistics, as in
Smith and Clark (1993), are based on a total n of less
than 20. The ANOVA tests reported below compare
both means for subjects, and for items.
86
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
Fig. 1. Stills illustrating a number of annotated visual features; the description and examples represent the marked settings for each
feature.
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
Results
It appeared that the participants found a majority of
the questions very easy, as they gave a maximum FOK
score of 7–61.1% of the questions, a score of 6–13.4% of
the questions, and lower scores to the remaining 25.5%. In
addition, 71.9% of the questions of the Wrst task were
indeed answered correctly and 89.6% of the same list of
questions in the multiple-choice test. Table 1 lists the average FOK scores as a function of Question Type (open
question versus multiple choice), and the response categories (correct answers, incorrect answers, and non-answers).
Speakers’ mean FOK ratings were higher when they
were able to produce an answer than when they were not
(with subjects as random factor, F1(1,17) D 71.821, p < .001;
with items as random factor, F2(1,23) D 59.028, p < .001).
The mean FOK ratings were higher for correctly recalled
answers than for the incorrect ones (F1(1,19) D 149.233,
p < .001; F2(1,35) D 38.086, p < .001). Speakers also gave
higher FOK ratings on average for responses that they
later recognized in the multiple choice test
(F1(1,18) D 55.018, p < .001; F2(1,27) D 19.714, p < .001).
These data thus show that there is a close correspondence
between the FOK scores and the correctness or incorrectness of a response in both the open test and the multiplechoice. The results are similar to those of Smith and
Clark (1993) and Brennan and Williams (1995).
In general, it can be observed that participants diVer in
the extent to which they use diVerent audiovisual features,
but these diVerences do not appear to relate to age or sex
of the participants. Table 2 lists the correlation coeYcients
between the FOK scores and number of words, gaze acts
(deWned as the sum of low, high, and diverted gaze) and
marked features (deWned as the sum of all features minus
words and gaze). It can be seen that there are negative correlations between the FOK scores and these variables for
answers, and positive correlations for non-answers. In
other words, for answers, higher FOK scores correspond
with a lower number of words, gaze acts and marked features, and the opposite relation holds for non-answers,
where higher FOK scores correspond with more words,
more gaze acts, and more marked features.
An analogous picture emerges from Tables 3 and 4,
which give the average FOK scores for presence versus
Table 1
Average FOK scores (and standard deviations) for diVerent
response categories
Experiment
Response
FOK
Open question
All answers (n D 704)
Correct answers (n D 575)
Incorrect answers (n D 129)
All non-answers (n D 96)
6.32 (1.27)
6.55 (1.00)
5.29 (1.72)
3.03 (2.12)
Multiple choice
Correct answers (n D 717)
Incorrect answers (n D 83)
6.17 (1.53)
3.84 (2.18)
87
Table 2
Pearson correlation coeYcients of FOK scores with words
(number of words), gaze acts (sum of low, high, and diverted
gaze), and marked features (sum of Wller, delay, high intonation,
eyebrow, smile, and funny face)
Correlations of
FOK scores with
Words
Gaze acts
Marked features
Response
Answers (n D 20)
¤¤¤
¡.265
¡.325¤¤¤
¡.410¤¤¤
Non-answers (n D 13)
.695¤¤¤
.630¤¤¤
.690¤¤¤
The n indicates for how many participants an individual correlation could be computed.
¤¤¤
p < .001.
absence of audiovisual features for answers and nonanswers, respectively. Table 3 shows that the presence of
a verbal feature (Wller, delay or high intonation) coincides with a signiWcantly lower FOK score. And similarly for the visual features, where the presence of an
eyebrow movement, a smile, a gaze act (high, low, and
diverted) or a funny face generally is associated with a
lower FOK score, which is signiWcant in all cases except
for the smile. As expected, the results in Table 4 display
the opposite pattern. When a verbal feature is present in
a non-answer this corresponds to a signiWcantly higher
FOK score for Wller and delay. Looking at the visual
cues, it can be seen that the presence of a marked feature
setting is associated with a higher FOK score as well,
albeit that the diVerences are only signiWcant for low and
high gaze, partly because of the limited number of data
points here (reXected by the lower n Wgures for nonanswers). Moreover, non-answers are perhaps inherently
less likely to be associated with a high intonation, presumably because speakers do not need to question their
own internal state (see, e.g., Geluykens, 1987), which is
reXected in a diVerence score of only 0.01.
To learn more about the cue value of combinations of
features, we also calculated, for answers and non-answers
separately, the average FOK scores for responses that
diVer regarding the number of marked feature settings
(minimum: 0, maximum: 7). Note that the theoretical
possibility of an utterance containing 8 or 9 marked feature settings did not occur in our data set. The results of
this are visualized in Fig. 2, which again illustrates opposite trends for the two response categories: for answers,
the average FOK score decreases with an increasing
number of marked features, while the opposite is true for
non-answers (cf. also the last row of Table 2).
Table 5 lists the distribution (in percentages) of cases
with a marked setting for our diVerent auditory and
visual features as a function of the relative number of
marked features present (1–7). First, looking at the n values in the left column, this table reveals that there are
considerable diVerences in relative frequency between the
diVerent features: the diVerent gaze acts (low, high, and
88
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
Table 3
Average FOK scores (and standard deviations) for answers as a function of presence or absence of audiovisual features
Auditory
Filler (n D 19)
Delay (n D 20)
High intonation (n D 19)
Visual
Eyebrow (n D 19)
Smile (n D 17)
Low gaze (n D 20)
High gaze (n D 19)
Diverted gaze (n D 20)
Funny face (n D 10)
Present (1)
Absent (2)
DiV. (1) ¡ (2)
5.70 (1.00)
5.23 (0.80)
5.91 (0.80)
6.53 (0.32)
6.53 (0.27)
6.40 (0.27)
¡0.83¤¤¤
¡1.30¤¤¤
¡0.49¤
5.78 (0.87)
5.69 (1.36)
6.05 (0.61)
5.72 (0.98)
5.96 (0.57)
4.78 (1.19)
6.46 (0.26)
6.36 (0.35)
6.47 (0.33)
6.52 (0.34)
6.64 (0.38)
6.38 (0.43)
¡0.69¤¤¤
¡0.67
¡0.42¤¤
¡0.80¤¤¤
¡0.68¤¤¤
¡1.60¤¤
Statistics are based on paired t test analyses. The n indicates the number of participants for which individual means could be computed.
The DiV. score is obtained by subtracting the average Absent FOK score from the average Present FOK score.
¤
p < .05.
¤¤
p < .01.
¤¤¤
p < .001.
Table 4
Average FOK scores (and standard deviations) for non-answers as a function of presence or absence of a audiovisual features
Auditory
Filler (n D 9)
Delay (n D 12)
High intonation (n D 5)
Visual
Eyebrow (n D 8)
Smile (n D 12)
Low gaze (n D 14)
High gaze (n D 10)
Diverted gaze (n D 9)
Funny face (n D 4)
Present (1)
Absent (2)
DiV. (1) ¡ (2)
4.98 (2.05)
4.21 (2.22)
3.67 (1.70)
2.60 (1.20)
2.20 (1.16)
3.66 (1.21)
+2.38¤¤
+2.01¤
+0.01
3.76 (1.99)
3.69 (2.33)
3.96 (2.05)
4.00 (1.75)
3.43 (1.44)
3.75 (2.63)
2.41 (0.92)
2.70 (1.13)
2.49 (1.21)
2.25 (1.50)
2.79 (2.14)
2.85 (1.44)
+1.35
+0.99
+1.47¤¤
+1.75¤
+0.64
+0.90
Statistics are based on paired t test analyses. The n indicates the number of participants for which individual means could be computed.
The DiV. score is obtained by subtracting the average Absent FOK score from the average Present FOK score.
¤
p < .05.
¤¤
p < .01.
especially diverted) and high intonation occur very often,
as opposed to the features smile and funny face which are
more rarely used. But, in addition, the table also makes it
clear that some features can relatively more often be used
in isolation or with only one more other feature present
(like, low gaze, smile, and high intonation), whereas features like funny face and especially delay tend to co-occur
with a high number of marked settings for other features.
Discussion
This Wrst study has replicated some of the Wndings of
the research by Smith and Clark (1993), and extended
them into the visual domain. It was found that our participants’ FOK scores correspond with their performance in
the open question and multiple-choice test, and, in
addition, particular audiovisual surface forms of the
utterances produced by our speakers are indicative of the
amount of conWdence speakers have about the correctness of their response. For answers, lower scores correlate
with occurrences of long delay, Wllers and high intonation,
as well as a number of gaze features, eyebrow movements,
funny faces, and smile. In addition, speakers tend to use
more words and more gaze acts, when they have a lower
FOK. For non-answers, the relationship between FOK
scores and the diVerent audiovisual features is the mirror
image of the outcome with answers, which is as expected
given that both high FOK answers and low FOK nonanswers correspond to speaker certainty. If we compare
the average FOK scores with the frequency of occurrence
for the various marked settings of audiovisual features, it
can be seen that the presence or absence of some highly
frequent features such as diverted gaze and high intonation has a relatively small impact on the average FOK
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
89
Materials
From the original 800 responses, we selected 60 utterances, of which 30 were answers and 30 non-answers. Of
both the answers and the non-answers, 15 had high FOK
and 15 low FOK scores. Following Brennan and Williams (1995), only the answer of a question–answer pair
was presented to participants, to avoid that participants
would unconsciously use their own estimation of the
question’s diYculty in their perception of the answer.
The selection was based on written transcriptions of
the responses by someone who had not heard or seen the
original responses. Given the individual diVerences in
the use of the FOK scale, we chose to use—per
speaker—the two highest scores as instantiations of high
FOK scores and the two lowest as low FOK scores. The
(in)correctness of the answer was not taken into account
when selecting the stimuli, hence both high FOK and
low FOK answers could be incorrect. We initially made
a random selection of stimuli meeting these restrictions,
but utterances were iteratively replaced until the following criteria were met:
Fig. 2. Average FOK scores for answers and non-answers as a
function of the relative number of marked prosodic features
(minimum D 0, maximum D 7; combinations of 8 or more features do not occur).
score, while this eVect is substantial for an infrequent feature such as funny face (especially for answers).
The Wrst study was a speaker-oriented approach to
gain insight into audiovisual correlates of FOK. While
our analyses revealed that there was a statistical relationship between the two, this in itself does not prove that
the audiovisual properties also have communicative relevance. To prove this, we performed a perception study,
for which we used earlier work by Brennan and Williams
(1995) as our main source of inspiration.
1. The original question posed by the experimenter
should not appear again in the participants’ response.
2. All the answers should be lexically distinct, and
should thus not occur twice. This criterion was not
applied to the non-answers as they were very similar.
3. The responses should be maximally distributed
across speakers. There should be maximally two
answers and two non-answers per speaker.
Experiment 2
Having applied this procedure on the basis of written
transcriptions of the data, we Wnally replaced a couple of
stimuli by others, if the signal-to-noise ratio made them
unsuitable for the perception experiment.
Method
Participants
Participants were 120 native speakers of Dutch, students from the University of Tilburg, none of whom had
participated as speaker in the Wrst experiment.
Procedure
The selected stimuli were presented to participants in
three diVerent conditions as a group experiment: one
Table 5
Percentage of marked settings for diVerent audiovisual features as a function of the relative number of marked prosodic features
Number of marked feature settings
Auditory
Filler (n D 191)
Delay (n D 161)
High intonation (n D 232)
Visual
Eyebrow (n D 154)
Smile (n D 98)
Low gaze (n D 280)
High gaze (n D 232)
Diverted gaze (n D 409)
Funny face (n D 36)
Total
1
2
3
4
5
6
7
6.3
0.6
13.8
15.7
5.0
20.7
18.8
17.4
25.0
17.8
24.8
15.5
18.8
23.0
10.8
13.6
17.4
8.2
8.9
11.2
5.6
100
100
100
11.0
15.3
22.1
3.9
12.5
0
15.6
17.3
22.5
21.1
24.9
8.3
14.3
10.2
18.2
21.1
20.5
11.1
22.7
17.3
10.0
20.7
17.4
19.4
14.9
17.3
12.1
13.4
12.0
22.2
12.3
14.3
8.6
12.5
8.1
16.7
8.4
7.1
6.1
6.9
4.4
19.4
100
100
100
100
100
100
Minimum D 1, maximum D 7; combinations of 8 or more features do not occur.
90
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
third of the participants saw the original clips as they
were recorded (Vision + Sound), another third saw the
same clips but then without the sound (Vision), whereas
the last third could only hear the utterances without seeing the image (Sound). In all three conditions, stimuli
were presented on a large cinema-like screen with loudspeakers to the right and left over which the sound was
played in the Vision + Sound and Sound condition.
Both the visual and the auditory information were
clearly perceptible for participants. Participants in each
condition were Wrst presented with the stimulus ID (1–
30) followed by the actual stimulus. In case of the sound
only stimuli, participants saw a black screen instead of
the original video recording. The motivation to present
sound only stimuli also visually was to make sure that
participants were aware of the start of the utterance, in
case there was a silent pause in the beginning of the
utterance. The interstimulus interval was 3 s. Within a
condition, participants had to judge recordings in two
separate sessions, one with answers as stimuli and one
with non-answers. The question to the participants
about the answers was whether a person appeared “very
uncertain” (1) or “very certain” (7) in his/her response.
The question for the non-answer stimuli was whether
participants thought the person would recognize the
correct answer in a multiple-choice test, with 1 meaning
“deWnitely not” and 7 “deWnitely yes.” Below, the scores
are referred to as the Feeling of Another’s Knowing
(FOAK) scores. Each part of the experiment was preceded by a short exercise session with two answers and
two non-answers, respectively, to make participants
acquainted with the kinds of stimulus materials and the
procedure.
Statistical analyses
The participants’ responses were statistically tested
with a repeated measures ANOVA with the FOAK
scores as dependent variable, with original FOK scores
and response type as within-subjects factors, and cue
modality (Vision, Sound, and Vision + Sound) as
between-subjects factor. Pairwise comparisons were
made with the Tukey HSD method.
Results
Table 6 shows that there were signiWcant eVects on the
participants’ FOAK scores of original FOK status of the
utterance (high FOK utterances receive higher FOAK
scores than low FOK utterances) and of response category (answers receive higher FOAK scores than nonanswers), while there was no main eVect of the cue
modality (Vision, Sound, and Vision + Sound). However,
there were signiWcant 2-way interactions between FOK
and cue modality (F(2,117) D 54.451, p < .001) and between
response and cue modality (F(1,117) D 241.597, p < .001),
and a signiWcant 3-way interaction between FOK, cue
modality and response (F(2,117) D 3.291, p < .05). Split
analyses showed that these main eVects and interactions
also hold when looking at the three cue modalities separately. For all cue modalities the diVerence in FOAK
score between high FOK and low FOK utterances was
signiWcant (Vision: F(1,39) D 270.070, p < .001; Sound:
F(1,39) D 957.515, p < .001; Vision + Sound: F(1,39) D
1516.617, p < .001), and similarly for response category
(Vision: F(1,39) D 16.629, p < .001; Sound: F(1,39) D 29.090,
p < .001; Vision + Sound: F(1,39) D 54.977, p < .001). There
was always a signiWcant interaction between FOK status
and response category (Vision: F(1,39) D 79.179, p < .001;
Sound: F(1,39) D 41.307, p < .001; and Vision + Sound:
F(1,39) D 161.703, p < .001).
The 2-way interactions can easily be understood
when we look at the average scores for combinations of
FOK and cue modality, and response and cue modality
(see Tables 7 and 8, respectively). The Wrst table shows
that the diVerence in scores for low and high FOK scores
Table 7
FOAK scores (and SE) for high FOK and low FOK stimuli in
diVerent experimental conditions
Cue modality
Vision
Sound
Vision + Sound
High FOK (1) Low FOK (2) DiV. (1) ¡ (2)
4.434 (0.061)
4.890 (0.061)
5.052 (0.061)
2.903 (0.061)
2.668 (0.061)
2.367 (0.061)
1.531
2.222
2.685
Table 6
ANOVA results (main eVects) for perception experiment: average FOAK scores as a function of FOK, response category, and experimental condition
Factor
Within subjects
FOK
Response
Between subjects
Cue modality
Level
FOAK (SE)
F stats
High
Low
Answer
Non-answer
4.792 (0.035)
2.646 (0.035)
3.922 (0.030)
3.516 (0.038)
F(1,117) D 2229.886, p < .0001
Vision
Sound
Vision + Sound
3.779 (0.047)
3.669 (0.047)
3.709 (0.047)
F(1,117) D 90.477, p < .0001
F(2,117) D 1.424, p D .245
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
Table 8
FOAK scores (and SE) for high FOK and low FOK stimuli for
answers and non-answers
Response
Answer
Non-answer
High FOK (1)
Low FOK (2)
DiV. (1) ¡ (2)
5.231 (0.044)
4.353 (0.045)
2.614 (0.035)
2.678 (0.051)
2.617
1.675
is more extreme in the Vision + Sound condition, than in
the unimodal conditions, meaning that the participants’
ratings were more accurate when participants had access
to both sound and vision. Notice that this explains why
no main eVect of experimental condition was found: the
diVerences in FOAK scores between high FOK and low
FOK stimuli change, while the overall FOAK averages
stay the same (see Table 6). The second table shows that
the diVerence between high and low FOK scores is—as
expected—easier to perceive in answers than in nonanswers. A separate analysis of variance revealed that
the diVerences between FOAK scores for high FOK and
low FOK stimuli were signiWcantly diVerent for the
diVerent cue modalities (F(2,237) D 46.248, p < .001). Post
hoc pairwise comparisons using Tukey HSD tests
showed that the diVerences in FOAK ratings between
high FOK and low FOK stimuli for a given cue modality
was signiWcantly diVerent from the diVerences in ratings
for any of the other two cue modalities.
Discussion
The results of the second perception test are consistent
with the Wndings of the Wrst analysis of speaker’s expression of uncertainty. It appears that participants are able to
diVerentiate low FOK responses from high FOK
responses in the unimodal experimental conditions, but
they clearly performed most accurately in a bimodal condition. This suggests that the addition of visual information, which was not present in the aforementioned FOK
and FOAK studies, is beneWcial for detecting uncertainty.
While we had seen that answers and non-answers exhibit
opposite audiovisual features, human participants are able
to adapt their judgments: they are able to tell the diVerence between low and high FOK for both response categories, be it that the performance for non-answers drops
compared to answers, in line with previous observations
of Brennan and Williams (1995). In conclusion, this study
brought to light that the audiovisual features of our original utterances have communicative relevance as they can
be interpreted by listeners as cues of a speakers’ level of
conWdence.
Since only high and low FOK stimuli were used in
Experiment 2, it would be interesting to see whether the
FOAK eVect persists for middle range FOK stimuli. We
hypothesize that participants would indeed note that
speakers are relatively uncertain when uttering a
medium FOK answer, since, generally, a lower FOK
91
answer corresponds to more audiovisual marked cues
(cf. Fig. 2), but how uncertain speakers are judged
remains an open question. Additionally, the current
experiment is a judgement study in which participants
are speciWcally asked to rate the perceived FOK of a
speaker’s utterance. It would be interesting to see if and
how people actually use audiovisual prosody as a cue for
FOK during natural conversations. We conjecture that
such cues will also be picked up and interpreted correctly
during ‘on line’ communication, and hope to address
this in future research.
General discussion
In response to a question, speakers are not always
equally conWdent about the correctness of their answer.
It has been shown that speakers may signal this using
verbal prosody (Brennan & Williams, 1995; Smith &
Clark, 1993). For instance, a response to the question
“Who wrote Hamlet?” might be “[pause] uh Shakespeare?” where the delay signals a prolonged memory
search, and the Wller and the high (question-like) intonation both signal relatively low conWdence from the metacognitive monitoring component.
In this paper, we have described two experiments
which extend these observations into the visual domain,
showing that there are clear visual cues for a speaker’s
feeling of knowing and that listeners are better capable
to estimate their feeling of another’s knowing on the
basis of combined auditory and visual information than
on the basis of auditory information alone. In the Wrst
experiment we used Hart’s (1965) paradigm to elicit
speaker utterances plus their accompanying FOK
scores. The data collected in this way were analyzed in
terms of a number of audiovisual cues. This analysis
conWrmed earlier work by Smith and Clark (1993) and
Brennan and Williams (1995) in that low FOK answers
are typically longer than high FOK answers, and tend
to contain more marked auditory cues (Wllers, longer
delay, and high intonation). Interestingly, we found that
the visual features show a comparable pattern. Focussing Wrst on answers, we found that marked facial
expressions (diverging from the ‘neutral’ face) such as
smiles, gaze acts, eyebrow movements, and ‘funny faces’
are more likely to occur in low than in high FOK
answers. In general, when a speaker gives a high FOK
answer to a question, this response is produced in a
straightforward way, and marked audiovisual cues are
not produced. When a speaker is uncertain of the
answer or has diYculty retrieving it, however, various
things may happen in parallel; low FOK answers tend
to require a longer search, which may trigger a Wller, but
may also be accompanied by various gaze acts. Both Wllers and gaze acts tend to occur during memory search,
before the actual answer is uttered. The gazing might
92
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
partly be explainable from the common observation
that it is easier to ‘think’ when not looking at a communication partner. We found that gazing typically
occurred in various directions both along the left-right
and up-down axis; in these data no single gaze act
appeared to fulWl a designated function. That a speaker
is uncertain about the correctness of a retrieved answer
can be indicated auditorily, via a rising, question-like
intonation, but also visually, for instance, by raising the
eyebrows (which has been interpreted as the visual
counterpart to a rising intonation, see, e.g., Bolinger,
1985). Similarly, the presence of a ‘funny face’ (often,
but not always, timed immediately after the answer has
been uttered) was also typically associated with a lower
FOK score, clearly functioning as a cue for speaker
uncertainty. Smiles, however, appeared to be more
ambiguous in this respect as some of our speakers
smiled when they were asked an ‘easy’ question for
which they had no diYculty retrieving the answer, while
others smiled in response to questions they considered
‘extremely’ diYcult or for which they felt they should
know the answer.
The earlier studies by Smith and Clark (1993) and
Brennan and Williams (1995) revealed an asymmetry
between answers and non-answers in that low FOK
answers look similar to high FOK non-answers (in both
cases the speaker is uncertain on a meta-cognitive level)
and that high FOK answers look similar to low FOK
non-answers, both in prosodic terms (in both cases the
speaker is certain). The current experiment conWrmed
this asymmetry, and extended it into the visual realm.
Thus, the presence of smiles, brow movements, gaze acts,
and a funny face is less likely for high FOK answers and
for low FOK non-answers. Data sparseness presumably
resulted in less signiWcant eVects for the non-answers
than for the answers (in addition, non-answers are lexically more similar than answers and are perhaps inherently less likely to end on a high boundary tone).
In the second experiment, participants watched
speaker utterances in one of three conditions (Sound
only, Vision only, and Vision + Sound) and had to judge
the speaker’s FOK. It turned out that overall participants are good at estimating the FOK of other speakers,
where the resulting FOAK scores are somewhat more
accurate for answers (high FOK answers receive high
FOAK scores and low FOK answers receive low FOAK
scores) than for non-answers. Interestingly, FOAK judgments are signiWcantly more accurate in the bimodal
(Vision + Sound) condition than in the respective unimodal ones. This indicates that the presence of visual information is actually beneWcial for FOAK judges. If we
compare the results for the Sound only and the Vision
only experiment, it appears that overall subjects made
better use of auditory than of visual cues for the perception of uncertainty. In terms of multisensory perception,
the relative dominance of auditory cues for FOK percep-
tion appears to be similar to the role of auditory cues for
information presentation.
However, it should be noted that visual expressions
such as funny faces and eyebrow movements occurred
relatively infrequently in the second experiment, but
when they did occur they seem to oVer a very strong cue
for FOAK estimations. Based on this observation we
recently performed an additional perception experiment, consisting only of incongruent stimuli combining
low FOK facial expressions (funny faces) with high
FOK speech prosody and vice versa. Participants were
asked how certain the speaker appeared in his or her
response. The results indicate that for these incongruent stimuli the facial expression in most cases has the
largest impact on the FOAK score, which is comparable to the aforementioned multisensory perception
results for emotion.
Acknowledgments
This research was conducted as part of the VIDIproject “Functions of Audiovisual Prosody (FOAP),”
sponsored by the Netherlands Organisation for ScientiWc Research (NWO), see foap.uvt.nl. Swerts is also
aYliated with the Flemish Fund for ScientiWc Research
(FWO-Flanders) and Antwerp University. Krahmer’s
work was carried out within the IMIX project “Interactive Multimodal Output Generation” (IMOGEN),
sponsored by the Netherlands Organisation for ScientiWc Research (NWO). Many thanks to Judith Schrier
(Antwerp) and Jorien Scholze (Tilburg) for their help
in carrying out the experiments. Many thanks also to
Pashiera Barkhuysen and Lennard van de Laar for
help with the annotation and for technical assistance,
to Annemarie Krahmer-Borduin, Per van der Wijst,
and Carel van Wijk for help with the statistics and the
test questions, and to the three anonymous reviewers
for their useful comments on a previous version of the
manuscript.
Appendix A
English translations of the Dutch questions used in the Wrst
experiment, as they were presented to participants in one of the
two random orders.
1.
2.
3.
4.
What do we call the sticks used in golf?
Who made the drawings for “Jip and Janneke”?
The sahara lies in which continent?
Which novel about a knight is the most reprinted book
after the Bible?
5. How many months does it take the moon to circle the
earth?
6. What does the abbreviation ‘Fl’ for the Dutch guilder
stand for?
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
7. What is the largest mammal?
8. What is the name of the gang of robbers that terrorized
Limburg in the 18th century?
9. Who, according to legend, was the bishop of Myra?
10. In which Dutch quiz show are the contestants awarded
with a toy monkey for each good answer?
11. What is the highest mountain of the Alps?
12. Who wrote Faust?
13. What is the chemical symbol of water?
14. What does the word ‘Jihad’ mean?
15. What color of light is used on the starboard side of a
boat?
16. What is Rembrandt’s last name?
17. Which television series is about the Forrester and Spectra families?
18. Guide Gezelle was a famous man. What was his occupation?
19. What is the boiling temperature for water?
20. In which wind-direction does one travel from Amsterdam to Brussel?
21. What is the name of the cartoon character who owns the
dog Pluto?
22. Egypt lies in which continent?
23. Who is the head of state of the Vatican?
24. Who wrote “The discovery of heaven”?
25. What is a “Friese doorloper”?
26. Brazil lies in which continent?
27. What is the pseudonym of the Mexican Don Diego de la
Vega?
28. Who wrote Hamlet?
29. Which rocker is also known as “The King”?
30. How many darts is a player allowed to throw in one turn?
31. In which wind-direction does one travel from London to
Berlin?
32. Which disease was known during the Middle Ages as
“The Black Death”?
33. What is the capital of Switzerland?
34. Supporters of which football club sing “Geen woorden
maar daden”?
35. How many degrees are in a circle?
36. Approximately, how many people live in the Netherlands and Belgium?
37. In which country did the Incas live?
38. Which person from the Bible went to look for mustard?
39. Which Dutch soap series has been running on television
for the longest period?
40. Who wrote the Iliad?
References
Andersen, J. (1999). Cognitive psychology and its implications
(5th ed). London: Worth Publishing.
Argyle, M., & Cook, M. (1976). Gaze and mutual gaze. Cambridge: Cambridge University Press.
Barkhuysen, P., Krahmer, E., & Swerts, M. (2005). Problem
detection in human-machine interactions based on facial
expressions of users. Speech Communication, 45, 343–359.
Benoît, C., Guiard-Marigny, T., Le GoV, B., & Adjoudani, A.
(1996). Which components of the face do humans and
machines best speechread?. In D. Stork & M. Hennecke (Eds.),
93
Speechreading by humans and machines: Models, systems, and
applications (pp. 315–328). New York: Springer-Verlag.
Bolinger, D. (1985). Intonation and its parts. London: Edward
Arnold.
Brennan, S. E., & Williams, M. (1995). The feeling of another’s
knowing: prosody and Wlled pauses as cues to listeners
about the metacognitive states of speakers. Journal of Memory and Language, 34, 383–398.
Bugenthal, D., Kaswan, J., Love, L., & Fox, M. (1970). Child
versus adult perception of evaluative messages in verbal,
vocal, and visual channels. Developmental Psychology, 2,
367–375.
Clark, H. (1996). Using language. Cambridge: Cambridge University Press.
Clark, H., & Fox Tree, J. (2002). Using uh and um in spontaneous speech. Cognition, 84, 73–111.
Clark, H., & Krych, M. (2004). Speaking while monitoring
addressees for understanding. Journal of Memory and Language, 50, 62–81.
Cruttenden, A. (1986). Intonation. Cambridge: Cambridge University Press.
de Gelder, B., & Vroomen, J. (2000). The perception of emotions by ear and by eye. Cognition and Emotion, 14(3), 289–
311.
Ekman, P. (1979). About brows: Emotional and conversational
signals. In M. von Cranach, K. Foppa, W. Lepenies, & D.
Plog (Eds.), Human ethology (pp. 169–202). Cambridge:
Cambridge University Press.
Ekman, P. (1999). Facial expressions. In T. Dalgleish & T.
Power (Eds.), The handbook of cognition and emotion (pp.
301–320). Sussex, UK: John Wiley & Sons Ltd.
Ekman, P., & Friesen, W. V. (1978). The facial acting coding system. Palo Alto: Consulting Psychologists’ Press.
Farah, M.J. (2000). The neural bases of mental imagery. In M.
Gazzaniga (Ed.), The cognitive neurosciences (2nd ed.). (pp.
965–974). Cambridge: MIT Press.
Frick-Horbury, D. (2002). The use of hand gestures as self-generated cues for recall of verbally associated targets. American Journal of Psychology, 115, 1–20.
Gale, C., & Monk, A. (2000). Where am I looking? The accuracy
of video-mediated gaze awareness. Perception & Psychophysics, 62, 586–595.
Geluykens, R. (1987). Intonation and speech act type. An experimental approach to rising intonation in declaratives. Journal of Pragmatics, 11, 483–494.
Glenberg, A., Schroeder, J., & Robertson, D. (1998). Averting
gaze disengages the environment and facilitates remembering. Memory & Cognition, 26, 651–658.
GoVman, E. (1967). Interaction ritual: Essays on face to face
behavior. Garden City, NY: Doubleday.
Goodwin, M. H., & Goodwin, C. (1986). Gesture and coparticipation in the activity of searching for a word. Semiotica, 62,
51–75.
GriYn, Z. (2001). Gaze durations during speech reXect word
selection and phonological encoding. Cognition, 82, 1–14.
Hart, J. T. (1965). Memory and the feeling-of-knowing experience. Journal of Educational Psychology, 56, 208–216.
Hess, U., Kappas, A., & Scherer, K. (1988). Multichannel communication of emotion: Synthetic signal production. In K.
Scherer (Ed.), Facets of emotion: Recent research (pp. 161–
182). Hillsdale, NJ: Erlbaum.
94
M. Swerts, E. Krahmer / Journal of Memory and Language 53 (2005) 81–94
Jameson, A., Nelson, T. O., Leonesio, R. J., & Narens, L. (1993).
The feeling of another person’s knowing. Journal of Memory
and Language, 32, 320–335.
Jordan, T., & Sergeant, P. (2000). EVects of distance on visual
and audio-visual speech recognition. Language and Speech,
43(1), 107–124.
Keating, P., Baroni, M., Mattys, S., Scarborough, R., Alwan, A.,
& Auer, E., et al. (2003). Optical phonetics and visual perception of lexical and phrasal stress in English. In Proceedings of the International Conference of Phonetic Sciences
(ICPhS) (pp. 2071–2074), Barcelona, Spain.
Kendon, A. (1994). Do gestures communicate? A review.
Research on Language and Social Interaction, 27, 175–200.
Kikyo, H., Ohki, K., & Miyashita, Y. (2002). Neural correlates
for feeling-of-knowing: An fMRI parametric analysis. Neuron, 36(1), 177–186.
Kohlrausch, A. & van de Par, S. (2005). Audio-visual interaction In Blauert, J. (Ed.), Communication acoustics. Heidelberg: Springer (in press).
Koriat, A. (1993). How do we know what we know. The accessibility account of the feeling of knowing. Psychological
Review, 100, 609–637.
Krahmer, E., & Swerts, M. (2004). More about brows. In ZS.
Ruttkay & C. Pelachaud (Eds.), From brows to trust: Evaluating embodied conversational agents (pp. 191–216). Dordrecht: Kluwer Academic Press.
Krauss, R., Chen, Y., & Chawla, P. (1996). Nonverbal behavior
and nonverbal communication: What do conversational
hand gestures tell us. Advances in Experimental Social Psychology, 28, 389–450.
Kraut, R., & Johnson, R. (1979). Social and emotional messages
of smiling. An ethological approach. Journal of Personality
and Social Psychology, 37, 1539–1553.
Ladd, D. R. (1996). Intonational phonology. Cambridge: Cambridge University Press.
Lawrence, B., Myerson, J., Ooknk, H., & Abrams, R. (2001).
The eVects of eye and limb movements on working memory.
Memory, 9, 433–444.
Massaro, D., & Egan, P. (1996). Perceiving aVect from the voice
and the face. Psychonomic Bulletin and Review, 3, 215–221.
McGurk, H., & MacDonald, J. (1976). Hearing lips and seeing
voices. Nature, 264, 746–748.
McNeill, D. (1992). Hand and mind: What gestures reveal about
thought. Chicago: Chicago University Press.
Mehrabian, A., & Ferris, S. (1967). Inference of attitudes from
nonverbal communication in two channels. Journal of Consulting Psychology, 31, 248–252.
Morgan, B. (1953). Question melodies in American English.
American Speech, 2, 181–191.
Morsella, E., & Krauss, R. (2004). The role of gestures in spatial
working memory and speech. The American Journal of Psychology, 117, 411–424.
Munhall, K., Jones, J., Callan, D., Kuratate, T., & VatikiotisBateson, E. (2004). Visual prosody and speech intelligibility.
Psychological Science, 15, 133–137.
Nelson, T. (1996). Consciousness and metacognition. American
psychologist, 51, 102–116.
Reder, L., & Ritter, F. (1992). What determines initial feeling of
knowing? Familiarity with question terms, not with the
answer. Journal of Experimental Psychology: Learning,
Memory and Cognition, 18, 435–452.
Sabbagh, M., & Baldwin, D. (2001). Learning words from
Knowledgeable versus Ignorant Speakers: Links between
preschoolers’ Theory of Mind and Semantic Development.
Child Development, 72(4), 1054–1070.
Smith, V. L., & Clark, H. H. (1993). On the course of answering questions. Journal of Memory and Language, 32, 25–
38.
Srinivasan, R., & Massaro, D. (2003). Perceiving prosody from
the face and voice: Distinguishing statements from echoic
questions in English. Language and Speech, 46, 1–22.
Sumby, W., & Pollack, I. (1954). Visual contribution to speech
intelligibility. Journal of the Acoustical Society of America,
26, 212–215.
Swerts, M., & Krahmer, E. (2004). Congruent and incongruent
audiovisual cues to prominence. In Proceedings of Speech
Prosody (pp. 271–274), Nara, Japan.
Walker, A., & Grolnick, W. (1983). Discrimination of vocal
expressions by young infants. Infant Behavior and Development, 6, 491–498.